Kaggle house price using some ML algorithm

Kaggle house price prediction

This is the sample R Markdown file following the workshop to create the template for Data Science workflow that we learnt inside class:

Step 1: Load library, Load data

rm(list=ls())
library(caret)

## Loading required package: lattice

## Loading required package: ggplot2

library(GGally)

## Registered S3 method overwritten by 'GGally':
##   method from   
##   +.gg   ggplot2

library(ggplot2)
library(RColorBrewer)
setwd("/home/tuev/Kaggle/house-prices")

df_train <- read.csv("https://raw.githubusercontent.com/vuminhtue/SMU_Data_Science_workflow_R/master/data/Kaggle_house_prices/train.csv",head=T)
df_test <- read.csv("https://raw.githubusercontent.com/vuminhtue/SMU_Data_Science_workflow_R/master/data/Kaggle_house_prices/test.csv",head=T)

Step 2: Select variables and check for missing value

df_small <- df_train[c("OverallQual","OverallCond","YearBuilt","X1stFlrSF",
                       "FullBath","GarageCars","SaleCondition","SalePrice")]
ggpairs(df_small,aes(colour=SaleCondition))

## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

sum(is.na(df_small))

## [1] 0

Convert SaleCondition to integer:

df_small$SaleCondition <- as.integer(as.factor(df_small$SaleCondition))

Step 3: Create partition for the data

ind1 <- createDataPartition(y=df_small$SalePrice,p=0.6,list=FALSE,times=1)
training <- df_small[ind1,]
testing  <- df_small[-ind1,] 

Step 4: Apply 1 ML algorithm to the data and calculate prediction

Apply Random Forest:

model_rf <- train(SalePrice~., data=training,method="rf",
                  prox=TRUE)
predict_rf <- predict(model_rf,testing)

Apply Neural Network

library(neuralnet)

smax <- apply(training,2,max)
smin <- apply(training,2,min)
trainNN <- as.data.frame(scale(training,center=smin,scale=smax-smin))
testNN <- as.data.frame(scale(testing,center=smin,scale=smax-smin))

set.seed(123)
ModNN <- neuralnet(SalePrice~.,trainNN, hidden=c(4,3),linear.output = T)
plot(ModNN)


predictNN <- compute(ModNN,testNN)
predictNN_rescaled <- predictNN$net.result*(smax-smin)[1]+smin[1]

Step 5: Evaluate the model output

Since the model output is continuous output, we use postResample to evaluate:

print(postResample(predict_rf,testing$SalePrice))

##         RMSE     Rsquared          MAE 
## 3.463785e+04 8.036627e-01 2.239034e+04

print(postResample(predictNN_rescaled,testing$SalePrice))

##         RMSE     Rsquared          MAE 
## 1.954434e+05 7.914847e-01 1.793110e+05

Display the scatter plots:

testing$predicted_rf <- predict_rf
ggplot(testing,aes(x=SalePrice,y=predict_rf))+
  geom_point(aes(color=factor(YearBuilt)))+
  geom_smooth(method="loess")

## `geom_smooth()` using formula 'y ~ x'

Since the Random Forest gave better results, we use this algorithm to predict the testing data:

Step 6: Predict the test data and write output

df_small_test <- df_test[c("OverallQual","OverallCond","YearBuilt","X1stFlrSF",
                       "FullBath","GarageCars","SaleCondition")]
df_small_test$SaleCondition <- as.integer(as.factor(df_small_test$SaleCondition))

output <- predict(model_rf,df_small_test)
output <- cbind(df_test$Id,output)

## Warning in cbind(df_test$Id, output): number of rows of result is not a multiple
## of vector length (arg 2)

colnames(output) <- c("ID","SalePrice")
write.csv(output,"Output.csv",row.names = FALSE)